A section of the results many clients dread is the 'don't know' column. A large portion of don't knows can ebb into the weight of consensus on an issue, and turn a solid, headline-generating majority into a disenchanting, disregarded plurality.
But a high proportion of don't knows can also tell us a lot, and can mask significant consensus among those who have an opinion.
Take for example, two questions from two different surveys. The first, a question to the public on whether directly elected police chiefs would help tackle corruption in the force; the second, a question to local councillors on whether they do or do not support government cuts to housing benefit.
Public: “Directly-elected police chiefs would be more likely to ensure that corrupt officers were brought to justice.”
Don’t know: 35%
Councillors: “I do not support the scale of government cuts to housing benefit.”
Don’t know: 9%
A majority of councillors agree with their statement, whereas a mere 2 in 5 members of the public agree that electing police chiefs would tackle corruption.
But look at the proportion of disagrees. A greater proportion of councillors disagree with the statement on government cuts than members of the public disagree with the statement on the police. Netting the agree / disagree scores illustrates the point.
“Directly-elected police chiefs would be more likely to ensure that corrupt officers were brought to justice.”
Net agree 15%
“I do not support the scale of government cuts to housing benefit.”
Net agree 13%
Therefore, the levels of agreement among those who state their opinion are virtually identical for both statements.
The proportion of ‘don’t knows’ can, therefore, tell us a huge amount about knowledge levels on the key issues among our participants. The real lesson from comparing these two questions is that the public are far less sure about the impact elected police chiefs will have on corruption, than local councillors are on their opinion of the Government’s cuts to housing benefit.
Of course, a high proportion of don't knows can also mean the question your pollster has drafted is unintelligible waffle instead of an articulate enquiry. In which case, it's probably best to shoot the messenger.